Configure Custom AI Providers with Model Management
wxrks' AI-powered features — CS Translation, Proofreading, Smells, Sous Chef, and more — already run on wxrks' own AI provider by default. Model Management is the account setting where you point some or all of those features at your own OpenAI, Azure, Anthropic, Groq, OpenRouter, Cerebras, or Google Gemini account instead, and control exactly which model handles each one.
💡 Who is this for? This guide is for Account Admins who need to configure custom AI providers and control which model handles each AI feature within the wxrks platform.
Prerequisites
Every wxrks account already includes AI tokens for the built-in AI features — Context Sensitive Translate, Smells, Tag Fixing, and the rest of wxrks Augmented Actions — at no extra setup. Model Management is only needed if you want to redirect those features to an AI provider account of your own.
Configuring your own provider requires the wxrks FUSION plan or higher. If your account isn't on a qualifying plan, Settings → Account Settings → Model Management shows an upgrade prompt instead of the settings list — contact your account manager or wxrks support to check or change your plan.
How Model Management is organized
Each entry you create in Model Management is called a row. A row pairs one AI provider configuration with the situations it should apply to. Three things decide when a given row is used:
Concept | What it controls |
Scope | Who the row applies to: your whole Account, or one specific Org Unit. |
Operations | Which AI features the row covers — e.g. CS Translation, Proofreading, Smells. Pick one or more, or leave this empty. |
Default row | A row with no Operations selected. It acts as a catch-all for any operation not explicitly covered by a more specific row at the same scope. |
You can create as many rows as you need — for example, one Account-wide default plus a few Org-Unit-specific overrides for teams that need a different model.
Getting to the page
Go to Settings → Account Settings → Model Management. You'll see the list of rows your account currently has — empty the first time you visit, since every account starts out using wxrks' built-in AI.
Create a Model Setting
Click Add Model Settings at the top of the page.
1. Choose a scope, name the row, and pick its operations
Scope defaults to Account. Switch it to Org Unit to scope this row to one specific org unit instead — a picker appears so you can select which one.
Name is just a friendly label so you recognize the row later in the list (e.g. “Claude via OpenRouter for translation”). Operations is where you pick which AI features this row covers — leave it empty to make the row a default for anything not otherwise covered at this scope.
2. Choose a provider
The Provider dropdown lists every AI provider wxrks currently supports:
Provider | What you need |
OpenAI | API key |
Azure OpenAI | API key, resource name, deployment name |
Anthropic | API key |
Groq | API key |
OpenRouter | API key |
Cerebras | API key |
Google Gemini (Studio) | API key |
Google Gemini (Vertex AI) | GCP project ID, region, service-account JSON |
Credentials are encrypted at rest — nobody at wxrks can read them back once saved. For how wxrks and these providers handle your content more broadly, see wxrks AI Privacy - Data Sharing with GPT Models.
3. Fill in the model and behavior fields
Once a provider is selected, the rest of the form appears:
Model Name — the specific model identifier, e.g.
gpt-4o-mini,anthropic/claude-3.5-sonnet,llama-3.1-70b-versatile.Base URL — auto-filled from the provider's default. Override only if you use a proxy or a regional endpoint.
Max Completion Tokens — the output-length cap per call. Leave it blank for the platform default of 4000; raise it for long-form generation, lower it for cost control.
Reasoning Effort shows up for providers and models that support it — it controls how much internal step-by-step reasoning the model does before answering. Higher effort can improve quality on difficult or ambiguous content, at the cost of speed and token usage. The exact choices available (from None up to Max) depend on the provider and model you've entered above; simpler models only offer Default.
Finally, fill in whatever credential fields the chosen provider needs (see the table above), and use the Enabled toggle if you want to save the row without activating it yet.
4. Save
Click Save. The row appears in the list immediately. Changes take effect within about a minute — wxrks caches routing decisions for performance, and saving a row evicts the relevant cache entries right away.
How wxrks decides which row to use
When an AI feature runs, wxrks looks for the most specific row that applies, in this order:
An Org Unit row for the exact operation, on the org unit the call came from.
An Org Unit default row (no operations picked) for that same org unit.
An Account row for the exact operation.
An Account default row (no operations picked).
The wxrks platform default, if none of your rows match.
Example
Say you set up three rows:
# | Scope | Operations | Provider |
A | Org Unit “Acme US” | Proofreading | Groq |
B | Account | CS Translation | OpenRouter (Claude) |
C | Account | (empty — default) | OpenAI GPT-4o |
What wxrks uses:
Proofreading in Acme US → row A (Groq).
Proofreading in any other org unit → row C (GPT-4o — nothing more specific matches).
CS Translation from any org unit → row B (OpenRouter).
Smells, or anything else, from any org unit → row C.
A small caveat
A handful of operations are managed centrally by wxrks for platform consistency. If one of these is in play, the platform's own setup wins regardless of what you've configured, and your row is effectively ignored for that operation. If an operation ever seems to ignore your configuration, contact wxrks support to confirm whether it's centrally managed on your release.
Custom Parameters for advanced users
Some providers accept extra options beyond the standard chat-completions fields. Custom Parameters (JSON) lets you pass those through as a raw JSON object, forwarded verbatim to the provider. It currently only appears for OpenAI, Groq, and OpenRouter rows.
Use case: pick which upstream OpenRouter uses
OpenRouter fronts many upstream providers. You can force it to use only one (say, Anthropic) with:
{ "provider": { "only": ["anthropic"], "allow_fallbacks": false, "data_collection": "deny" } }Use case: deterministic OpenAI output
{ "seed": 42, "response_format": { "type": "json_object" } }The field must be a valid JSON object — the Save button stays disabled until it parses. Anything the provider itself rejects comes back as an error on the AI call, not a save-time validation.
Editing, copying, and removing rows
From the list, each row offers:
Edit — click the row's name to reopen it pre-filled. Scope can't be changed on edit. Credential fields show blank; leave them blank to keep the existing values, or fill them in to replace them.
Copy — duplicates the row into a new one, and this is how you move a configuration to a different scope (e.g. turn an account-wide row into one scoped to a single org unit) without recreating it from scratch. Keep the same provider and you can leave the credential fields blank — wxrks reuses the original's encrypted credentials automatically. Switching provider requires entering new credentials for it.
Delete (trash icon) — removes the row after a confirmation prompt.
Troubleshooting
“Model Management shows an upgrade prompt instead of my settings.” Your account isn't on a plan that includes advanced Model Management — see Prerequisites above.
“I saved a row but the AI still seems to use the old model.” Routing decisions are cached for up to 10 minutes, but a save triggers an immediate eviction for the account or org unit it affects. Give it up to a minute; if it persists, confirm the row is Enabled and that its Operations and Scope actually match the call you're making.
“My Custom Parameters don't seem to do anything.” The field is only honored for OpenAI, Groq, and OpenRouter rows — it doesn't appear at all for other providers.
“The Save button is greyed out.” For Org Unit scope, make sure a unit is selected. Required credential fields must be filled in on create (on edit, blank fields keep the existing values). Custom Parameters, if used, must be valid JSON.
“Which org unit does this row belong to?” The list's Scope column shows the org unit's name under an Org-Unit-scoped row.
Related articles
wxrks Augmented Actions — what these AI features do and how to use them in the editor, and how to turn them on for an account or org unit.
wxrks Machine Translation Integration — for raw machine-translation engines (Amazon Translate, Google Cloud Translation, DeepL, and similar) rather than the AI/LLM providers covered here.
wxrks AI Privacy - Data Sharing with GPT Models — how wxrks and its AI providers handle your content.